DeepSeek opens Search and Fetch APIs at $0.75 per 1,000 requests

DeepSeek announced on 17 October 2026 that it is opening its in-house search engine to API customers. The change log and a new Search & Fetch guide describe two endpoints: Search, which returns structured results for a query, and Fetch, which returns a single web page as Markdown. Each costs $0.75 per 1,000 successful requests.

What the guide specifies

• Search (POST /v1/tools/search) takes a query of 1 to 128 characters and a result count from 1 to 50 (default 10). It supports site:, -site:, exact-phrase, lang:, after: and before: operators.
• Each result carries a title, URL, snippet, a 600-to-800-character summary meant for a model, a date, an authority tier from 0 to 4 and a relevance score from 0 to 3. Page content is returned only if need_content is true.
• Fetch (POST /v1/tools/fetch) takes a URL and returns the title and page content as Markdown. DeepSeek notes that fetching depends on the target site's availability and access policies, and a Fetch call can return 404 when a page is blocked or has no extractable content.
• Both use the same bearer-token authentication as the DeepSeek API and return OpenAI-style errors.

What it costs and what is capped

Billing is per successful request, not per token or per result. DeepSeek says a call is billable only when it returns HTTP 200 and produces a result: Search must return a non-empty array and Fetch must return content. Failed requests, empty searches and timeouts are not billed. At that rate 10,000 successful calls cost $7.50.

Search and Fetch are limited independently to 5 requests per second per account, which is 18,000 requests an hour at most on each. Fetch also has a per-domain limit that DeepSeek does not quantify, and requests over any limit get HTTP 429.

The guide adds one date. When you use the DeepSeek API through DeepSeek Harness, Claude Code or Codex, the server-side search and fetch calls already hit these APIs and are currently free. Per-request billing for them starts at 00:00 Beijing time on 1 November 2026, which is 16:00 UTC on 31 October. A team running an agent harness on DeepSeek models should assume its web-access calls will become a billed line from then.

What DeepSeek claims and does not say

The change log says the engine handles hundreds of millions of search requests a day, indexes hundreds of billions of pages and filters out SEO spam and low-quality machine-generated content. Those are DeepSeek's claims; the guide gives no quality benchmark, no comparison with other search APIs and no figure for freshness. The page I read does not say how long queries or fetched URLs are retained, or whether the index honours site operators' crawl rules. Both matter if an agent sends internal terms or customer URLs in a query.

What commenters said

The Hacker News thread (item 50138531; the site is not on this publication's source list, so it is not linked) had 184 points and 84 comments when I read it. Commenters compared prices from memory: one put Exa at $4 per 1,000, another Brave at $5 and Kagi at $12. I did not check those figures. The sharpest objection was practical: several asked how Fetch copes with bot protection such as Cloudflare checks or Anubis, and one noted the guide implies DeepSeek serves from its own index. Another asked whether DeepSeek pays the sites it fetches from. DeepSeek's page answers none of these.

What a team should do

• If your agents already search the web through DeepSeek Harness, Claude Code or Codex on DeepSeek models, count their search and fetch calls now and price them at $0.75 per 1,000 for the period from 1 November 2026.
• Test Fetch against the pages your agents actually read, including ones behind bot checks, before you rely on it. The guide promises no success rate.
• Decide what may appear in a query. Until DeepSeek states retention terms, keep customer names and internal identifiers out of agent search queries sent to this API.